On Sat, 4 Sep 1999, Jonathan Stone wrote:

> 
> In message <199909040851.SAA20151@balrog.supp.cpr.itg.telecom.com.au>
> Simon Burge writes:
> >Just a short note that I tried to get a pmax version of 'sdd' working,
> >but after two filesystem-related panics in less than an hour, I'm going
> >to give up :-(  If I can get a diskless setup working this week I'll
> >have another go, but I don't want to kill my main dev. box...
> 
> Was this onan ioasic machine? I think the Mi scsi code assumes

  I took this as Simon attempting to warp sdd to use the pmax raw SCSI
command interface.  I got the sources and had a quick glance at it, but I
haven't tried to modify it yet.

> it can do unbroken I/Os of more than 8k. Until we handle ioasic
> DMA-pointer reloads, or split larger transkactions into piecss <= 8k,
> very Bad Things will happen when the fs layer requests a read/write of more
> than 8k.
> 
> I have no idea whether Michael or Nisihmura-san's code handles this.

  The MI SCSI code does whatever the MD DMA routines tell it to do.  The
IOASIC version limits the DMA segment length to whatever will fit within
the two pages the IOASIC DMA pointers will do.  The MI SCSI layer will
keep starting DMA operations until all the data has been transferred or an
error occurs.  The TC option DMA code limits the DMA segment to 128K/7
(~17K).

Michael

---
Michael L. Hitch			mhitch@montana.edu
Computer Consultant
Information Technology Center
Montana State University	Bozeman, MT	USA